IMPROVING SENSOR MODEL ACCURAY USING MULTI-BAND AND MULTI-PATCH BASED MATCHING

초록

The accuracy of precise geometric correction of satellite images depends on the quality of Ground Control Point (GCP) data. To obtain high geometric correction accuracy with a limited number of GCP chips, we propose a method to acquire high quality GCP data using multi-band and multi-patch based matching. The proposed matching method works as follows. Firstly, we define initial search range in a satellite image using the initial sensor model. To define search range more precisely, we produce image pyramid of three reduced scale layers for the image and GCP chip. In this image pyramid, we performed single-band and single-patch matching between GCP chip and each image of the image pyramid. Secondly, we divide a multi-band GCP chip into six patches per each band: one patch with the original GCP chip size, one 2/3 size patch centered on the chip center and four 2/3 size patches at upper left, upper right, bottom left and bottom right of GCP chip. Finally, we adjusted search range of each patch precisely using the matching result of the image pyramid and performed multi-band and multi-patch matching. Through this matching process, we obtain 18 matching points per GCP chip. To verify the proposed method, we carried out experiments using KOMPSAT-3 and KOMPSAT-3A Level 1R pan-sharpened RGB 3-band images and GCP chips made from 0.25m Ground Sample Distance (GSD) aerial orthogonal images. We compared the proposed method with single-band and single-patch based matching. As a result of the experiment, the proposed method showed the best performance. When using this method, we could obtain the most inliers of matching points and the best accuracy of the sensor model.